The specification relates to speech recognition. In particular, the specification relates to a system for configuring custom vocabularies for speech recognition.
A user can issue a speech query to an automatic speech recognition system and receive a query result from the speech recognition system. However, the speech recognition system may have difficulty in recognizing some terms in the speech query correctly. For example, the speech recognition system may be unable to interpret a query for a place that is located near another location or an intersection. In another example, the speech recognition system may not recognize terms that have a personal meaning relevant to the user. Therefore, the query result received from the speech recognition system may not match the speech query or the system may be unable to interpret the query at all.
Existing systems may not account for whether the user is about to embark on a journey or query for direction information, and thus may not have the most pertinent or fresh user contact, location, or point of interest information available for use when recognizing user utterance.
In addition, some existing speech-based navigational systems are limited to using data stored locally on the device and do not include the most up-to-date or relevant data associated with the user. For instance, some systems only rely on a local contacts database and do not take into account the most recent communications that the user may have had, for instance, on a social network or via an instant messaging program. These systems also often do not account for the current geo-location of the user and whether the user's contacts or locations that the user is interested in are located near to that geo-location.
Some speech recognition systems may allow users to register a finite size of vocabulary in order to prevent the system from slowing down. However, if a large set of vocabulary is registered with these speech recognition systems, then the processing speed of speech recognition may be slowed down and the recognition data may possibly be corrupted with noisy data. As a result, the accuracy of these speech recognition system can be compromised.